Evaluation of Computed GDP-Figures

It is now of central importance to evaluate the estimated GDP-series. A first step would be to compute and compare GDP-values of the years 1998 to 2008 with the official GDP-figures that have been published since 1998.⁴⁰

40 There exist national account figures also for 2009 (including GDP), but they are still provisional and will be revised in December 2012, when the official national account figures for 2010 are going to be published by

TABLE 31:Computed and official GDP in the evaluation period (1998-2008)

TABLE 31 shows the results for the four components and their totals and contrasts them with the official GDP. The figures in italics denote computations of this chapter, while the values not in italics are official releases. When the official and the computed (nominal) GDP-figures are plotted in FIGURE 21, a close fit can be visually observed. The average absolute deviation from the official number during these eleven years is only 1.82%. This is low and clearly below the target by the national Office of Statistics, that calculates the official national accounts. The OFFICE OF STATISTICS [2012b, p.4] aspires a deviation of less than 3% of their published flash-estimate of GDP compared to the finally revised figures.

FIGURE 21:Evaluation of computed versus official GDP-figures

Also, a visual comparison of the nominal growth rates, which are very important to capture the business cycle pattern, unveils the good fit of the estimated series. In FIGURE 22, also the nominal GDP-estimates of the SGZZ, which were explained in the introduction of chapter 4., are displayed. As expected, those estimates are a bit lower than the estimations proposed here.

The author has carried out a similar technique to the one of the SGZZ for the years 1990-2008 and transposed the sectoral added value figures from Switzerland (which are publically ––––––––––––––––––––––––

the Office of Statistics. Also, the compiled data from the private companies only includes the time span 1972-2008. Therefore, only the years 1998-2008 are taken into account for this evaluation.

available from 1990 on) to Liechtenstein and has found out (as a-priori expected) that this strategy underestimated the actual GDP by about 10% to 25%. We therefore get the idea that the GDP values estimated here and those by the SGZZ correspond well with the impression of underestimation by the SGZZ-method.

FIGURE 22:Different sources of nominal GDP of Liechtenstein

We also observe again the close fit of the computed values with GDP-figures of the official national accounts.

Additional insights can be drawn from calculating the correlation of the four components of GDP (and of their total) with the official numbers, as shown in TABLE 32. The correlation coefficients of the estimated and the actual GDP are very high: 0.9975 in levels and 0.9605 for their annual percentage growth. Fortunately, the estimates of the gross operating surplus are very satisfying; this component has been the big challenge, since it contains a big share of GDP’s volatility and because there was no benchmark method for its generation, neither in the literature nor in the official national accounts of Liechtenstein (as they are calculated as residual from other accounts).

CORRELATION Compensation of Employees

Gross Operating

Surplus

Taxes on Production and Imports

Subsidies Aggregated Total (GDP)

Level 0.9915 0.9642 0.8912 0.9978 0.9975

%-Change 0.7676 0.8796 0.7769 0.9510 0.9605

TABLE 32:Correlation of estimated and actual figures (1998-2008)

The generated business cycle pattern seems very plausible, features no incomprehensible outliers and can be supported by economic interpretations (as has been done in the previous

sub-chapter). Also, the four sub-components of the GDP show no severe outliers, without interpretation by economic reasoning. They are displayed in the FIGURE 19andFIGURE 20.

If the two most important national accounting aggregates, which are now available in Liechtenstein over a longer time span, are compared then it is evident that both feature a very close cyclical pattern. Both time series, the aggregates real gross domestic product of Liechtenstein and real national income of Liechtenstein, were de-trended by the filter after HODRICK AND PRESCOTT [1997] in order to obtain a widely applied measure for the business cycle. The very high similarity of both backwardly estimated aggregates is evident.⁴¹ Hence, the computed GDP-figures from 1972 to 1997 are also plausible in this light.

FIGURE 23:Cyclical components of Liechtenstein’s real GDP and real national income

As observable from other countries (like Germany, Austria or Switzerland), the ratio of the compensation of employees to GDP has remained pretty stable over the past four decades.

Thus, it can serve as a good evaluation test to find out whether this is also true for the Liechtenstein. If this also holds for the estimated values from 1972 to 1997, then an additional hint is obtained that the estimated GDP-figures are plausible. FIGURE 24 shows the ratios for the four quoted countries. The ratio has remained between 0.5 and 0.6, during both periods 1972-1997 (estimated values) and 1998-2009 (official figures).

41 It is reasonable to compare both aggregates, which also imply approximated figures, for evaluations. Both series do not evolve from a similar econometric model and are not solely a main function of only few data sources or just dependent on a very similar computation method.

FIGURE 24:Ratio of Compensation of Employees relative to GDP

Also the ratio of the other very important component of GDP, the gross operating surplus, has been quite stable within a bandwidth of ranging from 0.35 to around 0.45 and shows no strong outliers.

FIGURE 25:Real growth rates of Liechtenstein’s and Switzerland’s GDP

It is interesting to see that the business cycle of Liechtenstein has very similar characteristics to the Swiss economic short-term fluctuations. This impression applies to both the growth rates of real GDP (see FIGURE 25) and the output gap measured as the percentage deviation from the long-run trend of GDP (HP-filter), as shown in FIGURE 26. Though, the amplitude is much higher, Liechtenstein is therefore characterized by a higher volatility.⁴²

42 Reasons could be the very high share of exports (due to the small size of the country): Exports are usually very volatile in the business cycle. Additionally, the smaller diversification across the production and provision of different goods and services compared to bigger countries may play an important role leading to higher volatility. As is has already been observed regarding national income also the gross domestic product featured no visible moderation of volatility over time.

FIGURE 26:Cyclical amplitude of real GDP in Liechtenstein and Switzerland

After having applied a trend-filter to Liechtenstein’s real GDP and inspected the cyclical behaviour (output gap), it is also worthwhile to concentrate on long-term growth (potential output) to compare the drift of the real GDP. A preliminary version of Liechtenstein’s GDP historic time-series is used that as well consists of the official values from 1998-2009 and computed values from 1972 until 1997 plus an econometric backward extension (1971-1954).

In that preliminary version the same computation method was executed via the same procedure as in the paper here, but as already mentioned, the figures were not the finally revised ones from this thesis. This preliminary time series was already applied in the contribution of BRUNHART, KELLERMANN AND SCHLAG [2012]. However, the preliminary estimated GDP-figures for 1972-1997 show almost completely the same cyclical pattern as the final ones in this paper. As the time series of Liechtenstein’s GDP in the just quoted paper was in addition econometrically extended by the author for the years 1971-1954 (applying a multiple regression model with indicators). These figures are now used in order to have a maximum number of observations. FIGURE 27 shows the growth of Liechtenstein’s potential output, this growth is also called trend drift (violet line in FIGURE 27, whereas the real percentage growth of the raw GDP is marked by the dashed blue line). The trend drift is measured by the percentage growth of the trend of real GDP derived by the method of HODRICK AND PRESCOTT [1997]. There are clearly observable three phases, the two main breaks being in the middle of the 70s and by the middle/end of the 90s.⁴³ Especially the interesting period for evaluation reasons (1972-1997) shows a very similar development to the growth of potential output in Switzerland (red line in FIGURE 27). This also supports the reasonability of the estimated GDP-values for Liechtenstein provided in this dissertation. Yet,

43 See also the remarks of footnote 24 in this context.

by the end of the sample, the continuous convergence between Switzerland and Liechtenstein seems to be completed.

FIGURE 27:Growth of potential output (drift) of Liechtenstein and Switzerland (Source: BRUNHART,KELLERMANN AND SCHLAG [2012, p.10])

Also, KELLERMANN[2005] has detected two main structural breaks in Switzerland’s economy for different important aggregates (such as GDP, GNI, production, employment) in the middle of the 70s and in 1997. This corresponds perfectly with the two observed structural breaks in the growth of potential output (drift), measured by the annual percentage growth of the trend (of Liechtenstein’s real GDP) that was determined applying the HP-filter on the newly estimated GDP-figures applying the method proposed in this dissertation (and already used in a preliminary version in BRUNHART,KELLERMANN AND SCHLAG [2012]). Hence, additionally to the findings considering the business cycle pattern, the newly generated GDP-figures have again an intuitive economic interpretation that supports their accuracy, also in the context of structural breaks in the long-term growth path.